HAL Id: hal-01649521https://hal.archives-ouvertes.fr/hal-01649521

Submitted on 7 Nov 2019

HAL is a multi-disciplinary open accessarchive for the deposit and dissemination of sci-entific research documents, whether they are pub-lished or not. The documents may come fromteaching and research institutions in France orabroad, or from public or private research centers.

L’archive ouverte pluridisciplinaire HAL, estdestinée au dépôt et à la diffusion de documentsscientifiques de niveau recherche, publiés ou non,émanant des établissements d’enseignement et derecherche français ou étrangers, des laboratoirespublics ou privés.

Continuous mixing of powder mixtures withpharmaceutical process constraints

Henri Berthiaux, Khadija Marikh, Cendrine Gatumel

To cite this version:Henri Berthiaux, Khadija Marikh, Cendrine Gatumel. Continuous mixing of powder mixtures withpharmaceutical process constraints. Chemical Engineering and Processing: Process Intensification,Elsevier, 2008, 47 (12), pp.2315-2322. �10.1016/j.cep.2008.01.009�. �hal-01649521�

Continuous mixing of powder mixtures withpharmaceutical process constraints

Henri Berthiaux ∗, Khadija Marikh, Cendrine GatumelRAPSODEE, UMR CNRS 2392, Ecole des Mines d’Albi-Carmaux, Campus Jarlard, route de Teillet, 81000 Albi, France

Abstract

While it would provide many advantages from many aspects, the application of continuous mixing processes to the pharmaceutical field is still in its infancy. In this paper we report results concerning the continuous mixing of nine ingredients (including three actives) that constitute a current drug. We examine these results in the light of different pharmaceutical process constraints, such as mixture quality control, time-stability of this quality, sensitivity of the process to perturbations. The apparatus is a pilot plant Gericke GCM 500 continuous mixer with three loss-in-weight feeders. A specific experimental protocol is developed to determine the homogeneity of the mixtures at the outlet of the mixer. The homogeneity of the mixtures is examined through industrial standards that would allow the product to be released on the market. The steady-state operation is first reported on, and it is demonstrated that a very acceptable mixture can be produced under certain conditions, with excellent time stability. The response of the mixer to filling sequences of two critical feeders is also quantified in terms of mixture homogeneity. It is found that it may be preferable to stop the process during these periods.

Keywords: Continuous mixing; Pharmaceuticals; Mixture quality; Powder; Transitory regime

1. Introduction

1.1. General industrial context with special attention to thepharmaceutical industry

Controlling powder-mixing operations is crucial in practi-cally all kind of industries: ceramics, agro-food, cement, aromas,explosives, cosmetics or drugs. By fixing the composition of amixture at a desired scale, it will largely guarantee the attainmentof the end-used properties of a product, even if, in most cases,this scale and/or these properties are not well defined. Chemi-cal and process engineers (and also product engineers) workingwith powders are currently faced with problems associated withmixture quality. They have to overcome such difficult barriersas sampling, use more or less advanced statistical analysis, copewith different standards and practices, but also understand a widerange of available technologies. Most of the time, when consid-ering the insertion of a new mixer in an existing process, costly

∗ Corresponding author.E-mail address: berthiau@enstimac.fr (H. Berthiaux).

pilot or full scale tests need to be performed. As a consequence,the general tendency in the industry is to avoid any change, andto concentrate on the way to validate the actual process. This isillustrated by a certain elasticity in sampling recommendationsand practices [1,2].

With regard to the pharmaceutical industry, the above gen-eral picture is even magnified. This is especially due totraceability needs and quality insurance. When transposed toa mixing problem, these requirements can be broken downinto:

• Qualification of the components. These must have been pre-viously characterised from the point of view of purity, particlesize, density, etc.

• Operational qualification regarding the training of the opera-tors.

• Qualification of the operations. Each operation in the processmust have been validated separately (mixer, feeders, etc.).Normally, this is the task of the equipment manufacturer.

• Process qualification is the responsibility of the producer.It must demonstrate that a process will give the productthe desired homogeneity, and may rely on optimal process

conditions, but also on sampling procedures and analyticalprotocols.

• Washing validation procedures. This is essential to avoidcross-contamination and generally concerns the analysis ofthe washing effluents.

This requires validation step-by-step, unit operation by unitoperation. It also means that a single change in a single opera-tion requires resetting the validation of the following steps, andsometimes of the previous ones. While this is essential for prod-uct quality and sanitary rules, it is a clear handicap for processinnovation in this field.

The “reconciliation principle”, which serves as the basis fortraceability, also forms part of such necessary but restrictingcontrols. This principle states that the flow of all the compo-nents of a mixture entering a process may be superior or equalto the flow leaving this process. In other words, if losses maybe explained, they cannot be found in another batch. Also, if amixture is found to be inhomogeneous, the whole related pro-duction must be destroyed. Even if there is clearly not a problemof product contamination by an external source, it will not bepermitted to mix the powders again. This is in contradiction withbasic process optimisation principles that are currently taught toyoung chemical engineers and demonstrates the main difficultythat a pharmaceutical company is currently faced with: how tostay competitive when process optimisation encounters so manyconstraints?

1.2. Mixture quality and pharmaceutical standards

First of all, the quality of a powder mixture cannot be definedif there is no precision on the scale at which the mixture isobserved. If this scale is the entire production, the whole batch,mixture quality is irrelevant. Conversely, if it is a single particleof a binary mixture, mixture quality will be zero. Normally,this scale of scrutiny corresponds to that of the unit dose that apatient may take. . . something between 10 mg and 5 g in typicalhuman recipes. When compared to the size and filling of anindustrial mixer, this means that it may contains from severalhundreds of thousands to one or two millions times the unitdose.

Secondly, one may identify what the component is for whichthe mixture has to be qualified. This key component is logi-cally the active ingredient, of therapeutic effect. But in the caseof multiple key components, what follows must be repeated foreach “active”. If a mixture is composed of N unit doses, thereforecorresponding to one batch or a definite period of time in con-tinuous operation, then mixture homogeneity will be expressedby the variance or standard deviation σ in the composition inactive xi of all the doses, with respect to the mean content µ:

σ2 = 1N

N∑

i=1

(xi − µ)2. (1)

However, due to the difference in scale explained above, andbecause in-line and non-destructive methods are not well devel-oped up to now, it is not feasible for a real pharmaceutical case to

use this formula at “full scale”. Sampling procedures are there-fore used to approach this criterion by taking n samples out ofthe N possible, thus defining:

s2 = 1n

n∑

i=1

(xi − xm)2 (2)

where xm is the mean content of active ingredient in the samples.Finally, because a standard deviation has a real significance ifit is compared to the mean value, the coefficient of variation isoften employed:

CV = s

xm. (3)

This overall analysis, including the sampling impact, servesas a basis for the definition of standards that need to be met toavoid the production being destroyed, with all the resulting costimplications. Basically three criteria are under observation:

• The mean of the samples xm. Even if there is always a sam-pling error, a significant difference from the “true” mean µ

can be indicative of a bad mixture. A typical criterion attachedto this is: µ − 7.5% µ < xm < µ + 7.5%µ.

• The individual values. A unit dose must contain the theoret-ical active composition with a certain tolerance. In practice:µ − 15% µ < xi < µ + 15%µ. So in a tablet containing 1 gof paracetamol, one may expect to have between 850 and1150 mg of the active.

• The coefficient of variation. It must be inferior to 6%. In pro-cess development, CV’s of around 1 or 2% are the objectivesto reach so as to limit the risk of homogeneity loss duringscale up.

Some differences can also be appreciated if one comparesUS Federal Drug Administration (FDA) and EU standards. Forexample, the rule cited above for the individual values is used inthe EU pharmacopoeia. However, the FDA does not mentionedthe true mean µ as the reference for calculating the range ofthe permitted values, which means that this can be done withestimated mean xm. In this case, the FDA standard is less severethat the EU one.

In parallel to the existence of these criteria for “batch libera-tion”, one may also be confronted with real practice. As statedabove, sampling must be done at the scale of scrutiny, whichmeans the scale of, say, one tablet. If this can be done eas-ily at the end of the process, it is not always feasible at thatprecise scale for the previous steps, including the mixing step,mainly because of sampler size and process accessibility (alsoresulting in higher sampling errors). This motivated the initiativeof developing Process Analytical Technologies (PAT), as on-line methods that are well adapted to detecting pharmaceuticalmolecules: NIR spectroscopy [3], FT-Raman [4], laser-inducedfluorescence [5,6].

Table 1Main characteristics of the powders used and mixing configuration

Mean particle size(!m)

Carr index (%) True specific gravity(g cm− 3)a

Theoretical weight perunit dose (g)

Mixing configuration Mass flows (kg/h)

BM containing A1and A2

110b 15 1.48 4.275 BM 32.87

A3 28c 21 1.22 0.025 A3 + 1/2 I1 0.95I1 67c 15 1.31 0.200I2 59c 11 1.33 0.050 I3 + I2 + 1/2 I1 4.18I3 53c 20 1.75 0.400

a Measured by Helium pycnometer.b Measured by sieving.c Measured by laser diffraction.

1.3. Batch versus continuous in the pharmaceuticalindustry

Recently, Pernenkil and Cooney [7] published a very com-plete review of continuous powder-mixing, some 30 years afterthe first one written by Williams [8]. They pointed out the littleattention that the scientific literature has paid to continuous pro-cesses for mixing powders and grains, particularly with respectto batch processes. They also noted the absence of reported workconcerning the continuous mixing of pharmaceutical powders.To our knowledge, the effective use of continuous mixers (as wellas continuous granulators) in pharmaceutical plants is restrictedto four or five examples throughout the world. Indeed, the batchreference predominates to an extent which is somewhat “cul-tural” in this field of activity, while in many cases, replacing anold batch mixer by a continuous one would result in a significantincrease in productivity.

Basic advantages of continuous mixers with respect to batchmixers are currently:

• Lower size of the mixing vessel for a same production level.• Less segregation risk due to the absence of handling opera-

tions, such as filling and emptying.• Lower running costs.• Better definition of mixture homogeneity, at the outlet of the

apparatus.

In the pharmaceutical context, we may add and emphasize:

• The possibility to include an on-line analysis set-up at theoutlet of the mixer to measure the quality of the mixtures, butalso to implement process control. This point is exactly in thedirect line of the PAT recommendations.

• The fact that practically all the final steps, such as tablettingand conditioning, in a drug fabrication scheme are alreadycontinuous operations.

• The elimination of scale-up problems during process devel-opment.

This last point is undoubtedly a very serious advantage forcontinuous mixers. The validation of an industrial “batch” dur-ing process development must actually be done at a scale of1/10 of the real batch capacity. This means that if one wants

Fig. 1. Pilot plant equipment (a) used showing the loss-in-weight feeders, andstirring device (b).

to produce 100 kg at industrial scale in a batch mixer, the val-idation can be done with a mixer containing 10 kg of mixture.In continuous mixing, this may be traduced by 1 h of full scaletest to represent 10 h of industrial production. The risk of erroris undoubtedly much easier to assume for a continuous processrather than for a batch process that has to “cross the scales”.

In this paper, we will report and discuss, probably for thevery first time, some results that we obtained when studyinga “real” pharmaceutical mixture of 9 ingredients including 3actives. First, we will present the continuous mixer used and thespecific methodologies developed. Then, we will focus on thehomogeneity of the mixtures produced and their time-stability innormal operation. To continue with process constraints, we willalso examine the effect of process disturbance, such as periodsof feeding of the loss-in-weight feeders.

2. Experiments

2.1. Mixture considered

The industrial case under consideration is a drug currentlyon the market containing 3 actives for a total of 9 ingredients.Two of the actives, which will be referred to as A1 and A2, areagglomerated with three other ingredients to form a basic mix-ture (BM). A1 will represent 10% (by weight) of the final drug,while A2 will concern 4% of it. The four last ingredients, threeadditives I1, I2, I3 and the active A3, are divided into two pre-mixes P1 and P2, which are defined in Table 1, as well as someother characteristics.

This finally defines a mixture made of three streams to mix:BM, P1, P2. The flow rates attached to these streams have beencalculated to cope with the industrial cadences of production,

Fig. 2. Photograph of the outlet of the mixer showing the sampling set-up.

and of course, with the composition of the mixture. As it can beseen from Table 1, the overall mixture is of low dosage concern-ing the active A3 (nearly 0.5%), which can be considered as the“main key component”. However, the mass of A3 in a sample ofunit size is still detectable by conventional methods. As regardsthe “physical” differences of the ingredients, the values do notreally indicate an important risk of particle segregation by sizeor density. We may only remark that flowability is worse for A3and I3, which are to be considered as fully cohesive powders andmay result in difficulties during dosage and mixing.

2.2. Mixing equipment and experimental procedure

The mixer used is a Gericke GCM 500® nearly hemi-cylindrical apparatus of the following dimensions: 50 cm long,

Fig. 3. Mixtures obtained after different operation times under the conditions specified on the graphs, showing the values of the individual content and the meancontent in A3, as well as the related coefficients of variation.

20 cm diameter (see Fig. 1). The stirrer consists of 15 bladesmounted on a driven shaft. The range of rotational speeds ofthe stirrer varies from 10 to 160 rpm, and may be advanta-geously described by a Froude number Fr = RN2/g, where Ris the radius of the mobile and N its rotational speed. In pre-vious studies [9–11], it has been shown that Froude valuesbelow 1 corresponded to dense phase flow, while Froude val-ues superior to 1 corresponded to nearly fluidised motion of theparticles.

Three loss-in-weight feeders of different capacities and char-acteristics ensure a very acceptable regularity and precision ofdosage. Qualification of these feeders for the present productswas made in a preliminary study, which is not reported here forclarity and confidentiality reasons.

To qualify the mixtures produced, a sampling protocol wasspecifically established. Striated boxes defining well separatedsections were placed on a conveyor belt of variable speed locatedat the outlet of the mixer (see Fig. 2). The speed of the beltwas adjusted so that the powder mass falling in a section cor-responded to the scale of scrutiny of the mixture (nearly 5 g).While a certain fluctuation exists of the sample masses collectedthrough this procedure, this error was neglected because theanalysis of mixture homogeneity is based on the concentrationsof the actives (and mainly A3).

For homogeneity calculations, 10 consecutive samples werecollected and the composition in each active component wascalculated after an industrially validated dissolution and HPLCspecific protocol. The three criteria defined in the introduc-tion of this paper were used to determine the best conditionsof operation of the mixer (stirrer type, rotational speed, pre-mix configuration). This “optimisation” procedure is not theobjective of the present work and will be reported in futurecommunications.

Fig. 4. CV and mean content in A3 obtained after 10 min of operation fordifferent rotational speeds.

3. Mixture quality in steady-state operation

Because the active A3 is low-dosed in the mixture, we willconcentrate our efforts on describing the homogeneity regardingthis component. Fig. 3 reports the results of the sampling pro-cedure described above for a stirrer rotational speed of 50 Hzand for several tries corresponding to different operation times.In the graphs, the tolerance intervals corresponding to the meanand to the individual values are also specified.

As can be seen from Fig. 3, no individual value is outsidethe tolerance interval, the sample mean lays in the acceptablerange and the CV values are much below the limit of 6%. After10 min running, which may be considered as corresponding tofive times the mean residence time of the particles, the mix-ture passes the three criteria for drug “liberation”. In addition,it seems that the mixture quality with respect to A3 improveswith the time of operation of the process. This may be due to

Fig. 5. Transitory regimes associated with a change in dosage mode.

a micromixing effect generated by the cohesive nature of thecomponents. A3 is a cohesive powder which is made of “pack-ets” of particles. These packets need a certain time to disruptinside the mixer and disperse into the bulk, therefore improv-ing the quality of the mixture. This may concern the particlesthat stay longer in the mixer, those belonging to the tail of aResidence Time Distribution curve. For instance, the micromix-ing steady-state characteristic time may be somewhat higherthan the macromixing time (the well-known “5 mean residencetimes” rule). This also suggests that a better premix of A3 withI1, would result in a better final mixture. For this kind of mixingproblem, there is also no doubt that a random sampling protocolwould have been very suitable for deriving the homogeneity ofthe mixtures.

In the above, the rotational speed was fixed to 130 rpm (or50 Hz for the engine), which corresponds to Fr = 1.89. In thisnearly fluidised regime, we also performed experiments witha lower and a higher rotational speed (see Fig. 4). While theother two values of the rotational speed (namely 40 and 60 Hz)still provided mixtures of industrially acceptable quality, theyalso gave rise to higher CV’s and mean values that were quitedifferent from the “hoped value” 25 mg. It may also be noted thatexperiments performed in the dense phase flow regime (below35 Hz), and not reported here, produced much worse mixturesand sometimes non-validated mixtures from the viewpoint ofthe present standards. Indeed, the choice (but also design) andthe operation of mixing equipment to resolve a specific mixingproblem is a very tedious task, as it still requires empiricism inthe optimisation procedure.

4. Impact of changes in dosage modes

Loss-in-weight feeders lose their regularity and precisionwhen the mass of powder in the hopper becomes less thanapproximately 15–20% of its apparent volume. This means thatsuch feeders may also be fed during processing, either through astorage tank with an aerolic conveyor or from a volumetric feederof high storage capacity. This results in an important source ofmixture homogeneity perturbation because during the time ded-icated to the filling of a loss-in-weight feeder, the powder massin the hopper changes too quickly to be counterbalanced by achange of speed of the feeding screw that may have been ableto ensure the same regularity and precision as before. In otherwords, a change of dosage mode, from gravimetric to volumet-ric, is operated and results in a transitory regime in the mixer(see Fig. 5). In this last part, we will examine how this problemaffects the quality of the mixtures under consideration.

The experimental protocol consists in: (a) measuring themixture quality after 10 min of normal operation; (b) feedinga loss-in-weight feeder for approximately t0 = 1 min; (c) mea-suring the mixture quality during the perturbation, after 1 meanresidence time (<t>); (d) measuring the mixture quality after5 mean residence times. Therefore, we assume that the mainimpact of the perturbation on the quality of the mixture corre-sponds to the powder mix produced at t0 + <t>.

Fig. 6 shows the three mixtures obtained before, during, andafter feeding of the feeder containing A3. As the flow rate asso-

Fig. 6. Mixtures obtained when filling the loss-in-weight feeder of active ingre-dient A3.

ciated with this feeder is quite low in comparison to the feedercontaining the other two actives, the mixture quality will still bejudged on the basis of the A3 content in the sample.

The mixture produced during this critical phase is not in con-formity with industrial standards, either from the viewpoint ofthe coefficient of variation or from the individual values, whilethe sample mean is still acceptable. Fortunately, there is a clearstability of the process, as the mixture examined after this phaseconforms to these criteria again. In particular, the value of theCV’s after and before feeding are extremely close to each other,and also to the value found in Fig. 3. This indicates a cer-tain reproducibility of the process, as well as demonstrating thesensitivity of our experimental approach.

Because this feeding operation is only to be repeated one ortwice during the whole production cycle, a possibility – giventhat it cannot be recycled – is to destroy the mixture producedduring this specific period of time. But if the cost of doing sois too high, the process may be stopped during re-feeding. Thisproblem arises again in Fig. 7, where the impact of similar exper-iments performed for the loss-in-weight feeder dedicated to BM(which contains A2 and A1) is shown for each of the three activeingredients. While A1 and A3 are practically insensitive to theperturbation, which is logical as A1 is the higher dosed compo-

Fig. 7. Changes in mixture qualities with respect to each active ingredient whenfeeding the loss-in-weight feeder containing the basic mixture.

nent and A3 is fed by another feeder, the coefficient of variationwith respect to A2 approaches 8% and the overall mixture is notvalidated.

Because the main flow rate is supported by this feeder, thissystematic perturbation will occur quite often (10–15 timesduring a production cycle). For instance, it may probably beadvisable to stop the process at a definite time of operation inorder to fill all the feeders. Fortunately, stop-and-go tests (notreported here) in which the mixer was stopped for between 1 minand several hours and then started again, have been shown to haveno influence on mixture quality. Nevertheless, an intermediatestorage of a small volume (a buffer) after the continuous mixerwill be necessary, at least for this reason.

5. Concluding remarks

In this paper, we have oriented the presentation of our resultstowards the “normal” operation of a continuous mixer operat-ing with pharmaceutical products, and examined the quality ofthe mixtures using industrial standards. We did not fully reportother results concerning the influence of operating parameters,the impact of accidental perturbations, the procedures for thebeginning and the end of the process, in particular to take account

of reconciliation principles. This will be done in future papers.Also, we may emphasise that the consideration of various keycomponents in the mixture is undoubtedly an additional diffi-culty: as in the present case, the risk is to invalidate the mixturebecause of one “indicator” out of nine.

While for the specific case under study, very acceptable con-ditions have been found to industrialise the process in the actualconfiguration, there are still many improvements to bring:

• The improvement of mixer design, at many levels: design ofthe inlet of the mixer to premix the ingredients in the chute;design of the outlet of the apparatus to adjust the mean res-idence time; design of the stirrer to approach the perfectlymixed vessel; . . .

• The inclusion of process constraints, such as facility of dosageof certain ingredients, during the formulation stage. It wouldbe a pity to give up so important a process optimisation schemeas changing from batch process to continuous process, justbecause a single additive was chosen rather than a differentone.

• The development of an on-line and real-time technique, witha fully validated associated methodology, for measuring thehomogeneity of the mixtures. This would in turn open thedoor to process control.

Acknowledgement

The authors wish to thank the company BMS (UPSA site inAgen, France) for their very fruitful scientific cooperation.

Appendix A. Nomenclature

CV coefficient of variation of a mixtureFr Froude numberg acceleration of gravity (m s− 2)n number of samples in an estimationN number of possible samples in a mixtureR radius of the stirrer (m)s2 variance of a mixture (estimated from sampling)t0 feeding time (s)<t> mean residence time (s)xi composition of sample i in key componentxm mean composition in key component (estimation)

Greek lettersµ mean composition in key component (real)σ2 variance of a mixture (estimated from sampling)

References

[1] F.J. Muzzio, P. Robinson, C. Wightman, D. Brone, Sampling practices inpowder blending, Int. J. Pharm. (1997) 155.

[2] J. Berman, A. Schoeneman, J.T. Shelton, Unit dose sampling: a tale of twothieves, Drug Dev. Ind. Pharm. 22 (11) (1996) 1121–1132.

[3] P.A. Hailey, P. Doherty, P. Tapsell, T. Oliver, P.K. Aldridge, Automatedsystem for the on-line monitoring of powder blending processes using near-infra-red spectroscopy. Part 1. System development and control, J. Pharm.Biomed. Anal. 14 (5) (1996) 551–559.

[4] G.J. Vergote, T.R.M. DeBeer, C. Vervaet, J.P. Remon, W.R.G. Baeyens,N. Diericx, F.F. Verpoort, In-line monitoring of a pharmaceutical blendingprocess using FT-Raman spectroscopy, Eur. J. Pharm. Sci. 21 (4) (2004)479–485.

[5] C.K. Lai, D. Holt, J.C. Leung, C.L. Cooney, G.K. Raju, P. Hansen, Real-time and non-invasive monitoring of dry powder blend homogeneity,AIChE J. 47 (11) (2001) 2618–2622.

[6] L. Pernenkil, T.P. Chirkot, C.L. Cooney, Continuous operation of zigzag®

blender for pharmaceutical applications, in: Proceedings of the WorldCongress on Particle Technology 5, Orlando, FL, 2006.

[7] L. Pernenkil, C.L. Cooney, A review on the continuous blending of pow-ders, Chem. Eng. Sci. 61 (2006) (2006) 720–742.

[8] J.C. Williams, Continuous mixing of solids, a review, Powder Technol. 15(2) (1976) 237–243.

[9] K. Marikh, Melange des poudres en continu: dynamique et modelisation,Ph.D., Thesis, INPL, 2003.

[10] K. Marikh, H. Berthiaux, V. Mizonov, E. Barntseva, D. Ponomarev, Flowanalysis and Markov chain modelling to quantify the agitation effect incontinuous powder mixer, Chem. Eng. Res. Des. 84 (A11) (2006) 1059–1074.

[11] K. Marikh, H. Berthiaux, V. Mizonov, E. Barantseva, Experimental studyof the stirring conditions taking place in a pilot plant continuous mixer ofparticulate solids, Powder Technol. 157 (2005) 138–143.